{
  "generatedAt": "2026-11-15T18:59:46+08:00",
  "site": "roy-tong.github.io",
  "count": 58,
  "dateSemantics": "每条的 date 是发表日期（实测类文章为评测/核数日期），不承诺产品此后的状态。复测以新文章追加，旧文不删。引用请带本条目的 url 与 date。",
  "disclosure": "https://roy-tong.github.io/disclosure/",
  "citationPolicy": "引用论点请引文章原文 URL（不是本索引）；使用前检查发表时间与更新的证据。",
  "posts": [
    {
      "contentId": "edge-ai-box-selection",
      "title": "端侧AI盒子怎么选：任务、替代方案与六个条件",
      "url": "https://roy-tong.github.io/notes/edge-ai-box-selection/",
      "date": "2026-11-06T00:40:00+08:00",
      "excerpt": "端侧AI盒子不是默认答案。用六个购买条件和五维框架，判断什么任务值得本地跑。",
      "tags": ["端侧 AI","硬件产品","评估框架"],
      "lang": "zh-CN"
    },
    {
      "contentId": "ai-glasses-feedback",
      "title": "38万条眼镜讨论，哪些能帮你做产品决定",
      "url": "https://roy-tong.github.io/notes/ai-glasses-feedback/",
      "date": "2026-11-06T00:30:00+08:00",
      "excerpt": "38万条AI眼镜反馈里，配网、续航、重量、处方镜片比 AI 能力更靠前。场景热度不等于需求。",
      "tags": ["AI 眼镜","用户研究","产品方法"],
      "lang": "zh-CN"
    },
    {
      "contentId": "agent-bill-verification",
      "title": "Verovo 费用/交付核对：从合成样例到可复现的验收协议",
      "url": "https://roy-tong.github.io/notes/agent-bill-verification/",
      "date": "2026-11-06T00:20:00+08:00",
      "excerpt": "Agent 账单不只有 token。用一份合成样例，拆开费用、交付、证据三个状态。",
      "tags": ["AI Agent","验收","产品方法"],
      "lang": "zh-CN"
    },
    {
      "contentId": "home-robot-task-closure",
      "title": "20万条机器人讨论，怎样变成产品判断",
      "url": "https://roy-tong.github.io/notes/home-robot-task-closure/",
      "date": "2026-11-06T00:10:00+08:00",
      "excerpt": "20万条机器人讨论能说明任务有没有跑通，不能直接算需求率。拆开分母、证据等级和场景差异。",
      "tags": ["具身智能","家庭机器人","产品研究"],
      "lang": "zh-CN"
    },
    {
      "contentId": "one-person-one-agent-three-products",
      "title": "One Person, One AI Agent, Three Products: I Wrote the Company into Code",
      "url": "https://roy-tong.github.io/en/notes/one-person-one-agent-three-products/",
      "date": "2026-10-31T08:30:00+08:00",
      "excerpt": "The question I get most often is not technical. It is arithmetic: “You’re one person. How do you run three products?”\n\n",
      "tags": ["agents","product method","working model"],
      "lang": "en"
    },
    {
      "contentId": "one-person-one-agent-three-products",
      "title": "一个人、一个 AI 代理、三个产品：我把公司写成了代码",
      "url": "https://roy-tong.github.io/notes/one-person-one-agent-three-products/",
      "date": "2026-10-31T08:20:00+08:00",
      "excerpt": "总有人问我：一个人怎么做三个产品？答案不是加班，是分工——代理当参谋，内核当执行，我当闸门。纪律不靠自觉，靠代码。",
      "tags": ["Agent","产品方法","工作模型"],
      "lang": "zh-CN"
    },
    {
      "contentId": "askovo-vs-your-research-stack",
      "title": "Askovo vs. Your Current Research Stack",
      "url": "https://roy-tong.github.io/en/notes/askovo-vs-your-research-stack/",
      "date": "2026-10-31T08:10:00+08:00",
      "excerpt": "AI research tools produce reports where every sentence sounds equally confident. The grounded and the invented look identical. If you sign decisions on that, you’re absorbing risk you can’t see — o...",
      "tags": ["user research","evidence","AI","tools"],
      "lang": "en"
    },
    {
      "contentId": "direction-map-2026",
      "title": "2026 探路地图：我在跟踪的十二个方向",
      "url": "https://roy-tong.github.io/notes/direction-map-2026/",
      "date": "2026-10-11T13:00:00+08:00",
      "excerpt": "一年研究、十二个方向，每个方向插一面旗：判断是什么、证据到哪、下一步验证什么。能走的路和走不通的崖，都在这张图上——插错了，我会回来拔。",
      "tags": ["研究盘点","方向判断"],
      "lang": "zh-CN"
    },
    {
      "contentId": "camera-robot-graveyard-chronicle",
      "title": "摄影机器人坟场编年：众筹神话的另一面名单",
      "url": "https://roy-tong.github.io/notes/camera-robot-graveyard-chronicle/",
      "date": "2026-10-11T12:00:00+08:00",
      "excerpt": "Beni 众筹 9169% 的同一年，Roll Cam 官网空白、Sporttotal 破产、MOVE'N SEE 被清算。把 2022 到 2026 的死伤排成编年史，死法高度雷同：把技术演示的掌声当成了需求证据。",
      "tags": ["摄影机器人","消费硬件","产品研究"],
      "lang": "zh-CN"
    },
    {
      "contentId": "who-defines-the-outcome",
      "title": "AI 按结果收费，「结果」是谁定义的？",
      "url": "https://roy-tong.github.io/notes/who-defines-the-outcome/",
      "date": "2026-10-11T11:00:00+08:00",
      "excerpt": "Salesforce 用 36 亿美元买下按解决收费的 Fin，买的不是模型，是对「解决」二字的定义权。从电表到话单再到 Agent 计费，新定价单位的生命周期从来没变过——现在轮到「结果」了。",
      "tags": ["Agent","商业化","产品研究"],
      "lang": "zh-CN"
    },
    {
      "contentId": "robot-gold-rush-picks-and-shovels",
      "title": "不做机器人的机器人公司：掘金潮里的卖水人",
      "url": "https://roy-tong.github.io/notes/robot-gold-rush-picks-and-shovels/",
      "date": "2026-10-11T10:00:00+08:00",
      "excerpt": "FieldAI 估值二十亿美元不造一台机器人，Gecko 靠爬墙机器拿到海军合同、值钱的是爬完墙之后的数据。把机器人行业的卖水人分成四层，你会发现价值正在离开身体。",
      "tags": ["机器人","具身智能","产品研究"],
      "lang": "zh-CN"
    },
    {
      "contentId": "when-product-scope-shrinks",
      "title": "产品越写越清楚，为什么离用户越来越远",
      "url": "https://roy-tong.github.io/notes/when-product-scope-shrinks/",
      "date": "2026-10-10T10:00:00+08:00",
      "excerpt": "一个便于验证的场景，怎样悄悄变成了整个产品的定义。运动影像方案的版本变化，留下了一个值得记住的教训。",
      "tags": ["产品研究","研究方法"],
      "lang": "zh-CN"
    },
    {
      "contentId": "research-labels-arent-users",
      "title": "二十万条讨论，为什么还不能告诉你谁会买",
      "url": "https://roy-tong.github.io/notes/research-labels-arent-users/",
      "date": "2026-10-10T10:00:00+08:00",
      "excerpt": "一条提交代码的 PR，经过数据管线，变成了用户求助。问题藏在标签增加的那一层意思里。",
      "tags": ["产品研究","研究方法"],
      "lang": "zh-CN"
    },
    {
      "contentId": "one-match-two-video-products",
      "title": "同一场比赛，为什么能卖出两种录像",
      "url": "https://roy-tong.github.io/notes/one-match-two-video-products/",
      "date": "2026-10-10T10:00:00+08:00",
      "excerpt": "教练找站位，家长找孩子。同一份比赛素材，为什么需要不同的产品组织方式？",
      "tags": ["产品研究","运动影像"],
      "lang": "zh-CN"
    },
    {
      "contentId": "research-review-2026",
      "title": "2026 研究盘点：已经站稳的判断，和还在等证据的判断",
      "url": "https://roy-tong.github.io/notes/research-review-2026/",
      "date": "2026-10-10T00:00:00+08:00",
      "excerpt": "一年研究、十二个方向，反复出现同一个越级模式。这篇盘点只用两类材料：已公开发布、可以点开核对的研究，和明确标注的研究讨论。十三个判断，每个都带证据和边界。",
      "tags": ["研究盘点","产品研究"],
      "lang": "zh-CN"
    },
    {
      "contentId": "ai-product-value-loop",
      "title": "AI 产品怎样从“做得到”走到“值得用”：采用、交付与验收的三道门槛",
      "url": "https://roy-tong.github.io/notes/ai-product-value-loop/",
      "date": "2026-10-10T00:00:00+08:00",
      "excerpt": "把拍摄、端侧 AI、创作软件与 Agent 计量放在一起，区分问题存在、方案采用、完整交付和可验证结果，并给出下一轮实验方法。",
      "tags": ["AI产品","产品研究","Agent","智能硬件"],
      "lang": "zh-CN"
    },
    {
      "contentId": null,
      "title": "从 90 条线索到 63 条核心样本：我是怎么做一张创业地图的",
      "url": "https://roy-tong.github.io/notes/startup-map-data-methodology/",
      "date": "2026-10-02T17:00:00+08:00",
      "excerpt": "第一个口径陷阱在单位：条，还是家。8 月 31 日我发了《一疆落，万物生》，讲大疆前员工创业地图；那篇的统计单元从头到尾是「条」，一条可以是公司、品牌或创业项目，所以这篇说 63 条核心样本，不说 63 家公司。没讲的部分这篇补上：90 条怎么来、凭什么砍到 63 条、哪些数我算了、哪些我拒绝算。\n\n",
      "tags": ["数据方法","创业生态","大疆"],
      "lang": "zh-CN"
    },
    {
      "contentId": null,
      "title": "消费级脑机接口的需求研究怎么做：消融、聚类和付费证据",
      "url": "https://roy-tong.github.io/notes/bci-demand-research-method/",
      "date": "2026-10-02T16:00:00+08:00",
      "excerpt": "在用户还没有稳定产品心智的新品类里，问卷和焦点小组很容易测到想象，而不是行为：问卷很难问出用户没经历过的东西，焦点小组收获的多半是对科幻的想象。消费级脑机接口属于”用户还没用过、但相关的痛与替代行为已经存在”的新品类，证据需要到用户已经留下痕迹的地方去找。这套研究我做过一轮完整落地，下面是复盘定型后的设计模式，七个环节，每个环节都有可审计的工件。\n\n",
      "tags": ["脑机接口","用户研究","研究方法"],
      "lang": "zh-CN"
    },
    {
      "contentId": null,
      "title": "「没有竞品」是最危险的竞争分析结论",
      "url": "https://roy-tong.github.io/notes/no-competitor-most-dangerous-claim/",
      "date": "2026-10-02T15:00:00+08:00",
      "excerpt": "第一份竞争报告的结论是：这个品类没有人占据。第二份报告把它推翻了——TareCount 就在做我想做的事，同一个锚点案例、同一套本地运行方案，明码标价 2500 美元一次完整审计。两份报告我都没有删：合在一起比任何一份单独看都值钱，前一份记录我怎么在铺好的搜索面里穷尽，后一份演示这个搜索面怎么漏、一个写过头的结论怎么被抓出来、写回方法。\n\n",
      "tags": ["研究方法","竞争分析","AgentMeasure"],
      "lang": "zh-CN"
    },
    {
      "contentId": null,
      "title": "Agent 时代的工作区，需要「唯一上游」",
      "url": "https://roy-tong.github.io/notes/agent-workspace-single-source/",
      "date": "2026-10-02T14:00:00+08:00",
      "excerpt": "2026-09-21 我盘点 ~/Documents/AI app/，du 报 28G；到 10-02 第二轮清理结束，同一个目录 22G。省下的 6G 是最不重要的数字。盘点让我看清的是另一件事：三个 AI 端（codex、ZCode、deepseek-harness）各自为政地在同一个工作区里干活，文件的产生机制跟人写代码的时代完全不一样，而我的工作区结构从没为此做过设计。这篇记录六类症...",
      "tags": ["Agent","工程实践","工作流"],
      "lang": "zh-CN"
    },
    {
      "contentId": null,
      "title": "我们审计了 110 个 AI 用量工具：数字都在哪里出错",
      "url": "https://roy-tong.github.io/notes/ai-usage-metering-audit-zh/",
      "date": "2026-10-02T13:00:00+08:00",
      "excerpt": "AI 产品的计费正在从包月转向按量与按结果，每个数字都出自一只表（meter），表多半握在厂商手里。过去一个月，我们审计了开源侧数 token、算成本、管预算的 110 个 AI 用量工具。核心判断先给：45+ 个已验证 bug，收敛成五个反复出现的家族；23 个修复已合并进上游，其中包括 34k star 的 langfuse；一位外部开发者复现了全部 236 项检查，还量化出一条路径把用...",
      "tags": ["Agent","AgentMeasure","计量","开源"],
      "lang": "zh-CN"
    },
    {
      "contentId": null,
      "title": "我查了 13 家研究机构，还是找不到全球第 4 名运动相机品牌",
      "url": "https://roy-tong.github.io/notes/action-camera-no4-untraceable/",
      "date": "2026-10-02T12:00:00+08:00",
      "excerpt": "IDC 说，2025 年全球手持智能相机出货 1665 万台、同比增长 83%（IDC 官网新闻稿，电头 2026-03-21）。前三名给得清楚：大疆 1,040 万台、62.4%，影石 340 万台、20.4%，GoPro 180 万台、10.8%（IT之家转引 IDC 图表，2026-03-21）。那么第 4 名是谁？我把公开渠道查了个遍，13 家机构、五个语种——至少在公开渠道里，ID...",
      "tags": ["智能硬件","市场数据","研究方法"],
      "lang": "zh-CN"
    },
    {
      "contentId": null,
      "title": "跑了几百个 Agent 会话后，我留下了三条系统规则",
      "url": "https://roy-tong.github.io/notes/agent-sessions-three-system-rules/",
      "date": "2026-10-02T11:00:00+08:00",
      "excerpt": "九月上旬某天，我让自建的 DeepSeek harness（下称 DSH）跑一轮竞品信息扫描，几十个子代理正在各自抓资料，其中一个突然发现内置搜索工具开始返回 HTTP 402——配额用完了。它没有停下来等我，自己翻了一遍工作区里的旧脚本，找到两条不花钱的检索路径，接着把自己那部分任务跑完了。\n\n",
      "tags": ["Agent","自动化","工程实践"],
      "lang": "zh-CN"
    },
    {
      "contentId": null,
      "title": "摄影机器人正在爆发，但还没人回答：为什么相机一定要会走？",
      "url": "https://roy-tong.github.io/notes/camera-robots-why-must-walk/",
      "date": "2026-10-02T10:00:00+08:00",
      "excerpt": "突然变热：一边创纪录，一边官宣不存在的产品\n\n",
      "tags": ["智能硬件","机器人","摄影","行业判断"],
      "lang": "zh-CN"
    },
    {
      "contentId": null,
      "title": "AI 应用的下一站，不是更多 Agent",
      "url": "https://roy-tong.github.io/notes/ai-applications-from-software-to-autonomy/",
      "date": "2026-09-07T11:30:00+08:00",
      "excerpt": "最近和一些投资人聊 AI 应用，得到的反馈有点互相打架。\n\n",
      "tags": ["AI 应用","Agent","AI Native Service","具身智能","世界模型","产业判断"],
      "lang": "zh-CN"
    },
    {
      "contentId": null,
      "title": "本地 AI 盒子，得先证明自己值得多占一个插座",
      "url": "https://roy-tong.github.io/notes/local-ai-box-task-economics/",
      "date": "2026-08-31T00:10:00+08:00",
      "excerpt": "英伟达把 DGX Spark 摆上桌面的时候，参数讲得很清楚：GB10 芯片、128GB 统一内存、1 PFLOP 的 FP4 稀疏算力、240W 功耗，3,999 美元。一台「个人 AI 超级计算机」，大小接近一台Mac mini 的放大版。但对于准备下单的用户，真正的问题藏在参数表之外：家里或办公室，是否愿意为它多留一个插座、一段持续在线的电费、一份固件更新的责任。\n\n",
      "tags": ["本地AI","智能硬件","产品策略"],
      "lang": "zh-CN"
    },
    {
      "contentId": null,
      "title": "AI 视频软件的护城河，藏在第二次修改里",
      "url": "https://roy-tong.github.io/notes/ai-video-second-edit/",
      "date": "2026-08-31T00:10:00+08:00",
      "excerpt": "给一个视频工具布置这样的修改任务：\n\n",
      "tags": ["AI视频","创作工具","产品判断"],
      "lang": "zh-CN"
    },
    {
      "contentId": null,
      "title": "一疆落，万物生",
      "url": "https://roy-tong.github.io/notes/dji-talent-whale-fall/",
      "date": "2026-08-31T00:00:00+08:00",
      "excerpt": "2026年8月19日，宇树科技在科创板上市，首日收盘市值3417亿元。\n\n",
      "tags": ["硬件","大疆","创业生态","组织"],
      "lang": "zh-CN"
    },
    {
      "contentId": "agentmeasure-from-spec-to-evidence",
      "title": "AgentMeasure Weekly: The Spec Is Growing Evidence",
      "url": "https://roy-tong.github.io/en/notes/agentmeasure-from-spec-to-evidence/",
      "date": "2026-08-25T14:20:00+08:00",
      "excerpt": "\n  The first AgentMeasure progress report. Project: github.com/roy-tong/AgentMeasure, website: roy-tong.github.io/AgentMeasure. Every number below comes from reproducible run output — no hand-trans...",
      "tags": ["agent","measurement","conformance","capability economy"],
      "lang": "en"
    },
    {
      "contentId": "agentmeasure-from-spec-to-evidence",
      "title": "AgentMeasure 一周进展：规范开始长出证据",
      "url": "https://roy-tong.github.io/notes/agentmeasure-from-spec-to-evidence/",
      "date": "2026-08-25T14:20:00+08:00",
      "excerpt": "\n  这是 AgentMeasure 的第一篇进展周报。项目主页：github.com/roy-tong/AgentMeasure，官网：roy-tong.github.io/AgentMeasure。以下所有数字都来自可复现的运行输出，不含手抄转写。\n\n\n",
      "tags": ["Agent","Capability Economy","计量","进展周报"],
      "lang": null
    },
    {
      "contentId": null,
      "title": "AI 应用不会被 Agent 塞满，它们会先被拆掉",
      "url": "https://roy-tong.github.io/notes/ai-apps-unbundled-before-agent-filled/",
      "date": "2026-08-21T10:00:00+08:00",
      "excerpt": "过去一年，AI 应用最常见的产品思路，是给现有软件加一个 Agent：CRM 加销售 Agent，文档工具加写作 Agent，设计软件加创作 Agent。\n\n",
      "tags": ["Agent","Harness","Capability Economy","SaaS","Ephemeral Software"],
      "lang": null
    },
    {
      "contentId": null,
      "title": "The Upsider 事件：Agent 第一次作为经济主体收付了钱",
      "url": "https://roy-tong.github.io/notes/the-upsider-event/",
      "date": "2026-08-18T20:30:00+08:00",
      "excerpt": "\n  事件事实以 @elliwoodtong 的 X 记录 为准（2026-08）：Upsider 是一个运行在 X 上的 AI Agent，它评估了与我的账户之间的交互，然后自动向我的链上地址发送了 token。本文区分事实与判断：事实标注来源，判断明确标注。\n\n\n",
      "tags": ["Agent","Capability Economy","计量","事件解读"],
      "lang": null
    },
    {
      "contentId": "agent-capability-monthly-01",
      "title": "Agent Capability Monthly · Issue 01: The Ecosystem Baseline",
      "url": "https://roy-tong.github.io/en/notes/agent-capability-monthly-01/",
      "date": "2026-08-18T09:30:00+08:00",
      "excerpt": "\n  Agent Capability Monthly is a fixed monthly report answering one question: as agents become software consumers, what is happening to the software economy. The structure is fixed in eight section...",
      "tags": ["agent","capability economy","monthly report","measurement"],
      "lang": "en"
    },
    {
      "contentId": "agent-capability-monthly-01",
      "title": "Agent Capability Monthly · Issue 01：生态基线",
      "url": "https://roy-tong.github.io/notes/agent-capability-monthly-01/",
      "date": "2026-08-18T09:00:00+08:00",
      "excerpt": "\n  Agent Capability Monthly 是每月一期的固定报告，回答一个问题：Agent 作为新的软件消费者，软件经济正在发生什么。结构固定为八段：新能力、新 provider、定价变化、MCP/Skills 生态、Agent 商业、计量进展、Top 5 信号、Roy 的判断。\n\n  创刊号是基线号：不追求当月新闻，而是把生态当前状态、计量方法与数据口径固定下来。后续每一期只报...",
      "tags": ["Agent","Capability Economy","月度报告","计量"],
      "lang": null
    },
    {
      "contentId": "every-agent-usage-number-is-self-reported",
      "title": "每个 Agent 用量数字，都是自报的",
      "url": "https://roy-tong.github.io/notes/every-agent-usage-number-is-self-reported-zh/",
      "date": "2026-08-17T00:45:00+08:00",
      "excerpt": "现场审计于 2026-08-16 完成，是 AgentMeasure Benchmark Run #001 的一部分。所有声称均真实、可溯源、可重放。\n\n",
      "tags": ["Agent","度量标准","Benchmark","Capability Economy"],
      "lang": null
    },
    {
      "contentId": "every-agent-usage-number-is-self-reported",
      "title": "Every Agent Usage Number Is Self-Reported",
      "url": "https://roy-tong.github.io/en/notes/every-agent-usage-number-is-self-reported/",
      "date": "2026-08-17T00:30:00+08:00",
      "excerpt": "Field audit conducted 2026-08-16 as part of AgentMeasure Benchmark Run #001. All claims are real, cited, and replayable.\n\n",
      "tags": ["agent","measurement","benchmark","capability economy"],
      "lang": "en"
    },
    {
      "contentId": "when-the-software-consumer-becomes-an-agent",
      "title": "When the Software Consumer Becomes an Agent",
      "url": "https://roy-tong.github.io/en/notes/when-the-software-consumer-becomes-an-agent/",
      "date": "2026-08-16T14:00:00+08:00",
      "excerpt": "The software consumer is changing from humans to agents. It is already happening — most people just haven’t yet treated it as something that deserves serious attention.\n\n",
      "tags": ["agents","capability economy","measurement","caas"],
      "lang": "en"
    },
    {
      "contentId": "when-the-software-consumer-becomes-an-agent",
      "title": "当软件的消费者变成 Agent",
      "url": "https://roy-tong.github.io/notes/when-the-software-consumer-becomes-an-agent/",
      "date": "2026-08-16T14:00:00+08:00",
      "excerpt": "软件的消费者正在从人变成 Agent。\n\n",
      "tags": ["Agent","Capability Economy","计量","CaaS"],
      "lang": null
    },
    {
      "contentId": null,
      "title": "How Software Usage by AI Agents Should Be Measured",
      "url": "https://roy-tong.github.io/en/notes/how-agent-usage-should-be-measured/",
      "date": "2026-08-16T10:00:00+08:00",
      "excerpt": "Whitepaper v0.2 · AgentMeasure Standard Draft 0.4\n\n",
      "tags": ["agent","measurement","capability economy","caas"],
      "lang": "en"
    },
    {
      "contentId": "agent-context-recommendation-after-rag",
      "title": "After RAG, Agents Need Context Recommendation",
      "url": "https://roy-tong.github.io/en/notes/agent-context-recommendation-after-rag/",
      "date": "2026-08-14T09:14:00+08:00",
      "excerpt": "The DeepSeek Harness architecture documentation contains one line:\n\n",
      "tags": ["AI agent","context engineering","DeepSeek Harness"],
      "lang": "en"
    },
    {
      "contentId": "agent-context-recommendation-after-rag",
      "title": "RAG 之后，Agent 需要 Context Recommendation",
      "url": "https://roy-tong.github.io/notes/agent-context-recommendation-after-rag/",
      "date": "2026-08-14T09:14:00+08:00",
      "excerpt": "DeepSeek Harness 的架构文档里有一句话：\n\n",
      "tags": ["AI Agent","Context Engineering","DeepSeek Harness"],
      "lang": null
    },
    {
      "contentId": null,
      "title": "机器人进家庭，先过“谁来收拾残局”这一关",
      "url": "https://roy-tong.github.io/notes/home-robots-harder-richer/",
      "date": "2026-08-14T09:13:00+08:00",
      "excerpt": "设想一个家庭机器人，能把桌上的杯子放回柜子。\n\n",
      "tags": ["家庭机器人","具身智能","产品判断"],
      "lang": "zh-CN"
    },
    {
      "contentId": "robots-give-people-new-bodies",
      "title": "Robots Don't Replace People — They Give People New Bodies",
      "url": "https://roy-tong.github.io/en/notes/robots-give-people-new-bodies/",
      "date": "2026-08-14T09:12:00+08:00",
      "excerpt": "I have worked on panoramic cameras, 360-degree drones, follow-cameras and consumer home-security robots. The hardware looks nothing alike — some you hold, some fly, some follow you around, some pat...",
      "tags": ["robotics","embodied intelligence","human-robot collaboration"],
      "lang": "en"
    },
    {
      "contentId": "robots-give-people-new-bodies",
      "title": "机器人不是替代人，而是给人增加新的身体",
      "url": "https://roy-tong.github.io/notes/robots-give-people-new-bodies/",
      "date": "2026-08-14T09:12:00+08:00",
      "excerpt": "我参与过全景相机、全景无人机和跟拍机，也做过家庭消费级安防机器人的产品工作。\n\n",
      "tags": ["机器人","具身智能","人机协作"],
      "lang": null
    },
    {
      "contentId": "appointed-manager-organizational-legitimacy",
      "title": "From Formal Authority to Organizational Legitimacy",
      "url": "https://roy-tong.github.io/en/notes/appointed-manager-organizational-legitimacy/",
      "date": "2026-08-14T09:11:00+08:00",
      "excerpt": "There is a management story making the rounds. The company and the person can’t be independently verified, so treat it as a management scenario, not a replicable success case.\n\n",
      "tags": ["organizational management","product leadership","team building"],
      "lang": "en"
    },
    {
      "contentId": "appointed-manager-organizational-legitimacy",
      "title": "空降管理者如何从职位合法性走向组织合法性",
      "url": "https://roy-tong.github.io/notes/appointed-manager-organizational-legitimacy/",
      "date": "2026-08-14T09:11:00+08:00",
      "excerpt": "我最近看到一个被反复转述的管理故事。\n\n",
      "tags": ["组织管理","产品领导力","团队建设"],
      "lang": null
    },
    {
      "contentId": "market-roles-remove-uncertainty",
      "title": "Geek, Professional, B2B, Consumer: Markets as Evidence Environments",
      "url": "https://roy-tong.github.io/en/notes/market-roles-remove-uncertainty/",
      "date": "2026-08-14T08:00:00+08:00",
      "excerpt": "Every new-product discussion hits the same two questions. B2B or consumer first? Then the familiar line appears:\n\n",
      "tags": ["market entry","product strategy","commercialization"],
      "lang": "en"
    },
    {
      "contentId": "market-roles-remove-uncertainty",
      "title": "Geek、Professional、B、C：不同市场负责消灭不同不确定性",
      "url": "https://roy-tong.github.io/notes/market-roles-remove-uncertainty/",
      "date": "2026-08-14T08:00:00+08:00",
      "excerpt": "讨论一个新产品怎样进入市场，会议里经常很快出现两个问题。\n\n",
      "tags": ["市场进入","产品策略","商业化"],
      "lang": null
    },
    {
      "contentId": null,
      "title": "AI 交互在等它的“触控时刻”",
      "url": "https://roy-tong.github.io/notes/ai-interaction-grammar/",
      "date": "2026-08-13T21:40:00+08:00",
      "excerpt": "打开一份 AI Wearable 竞品表，第一列通常是眼镜、耳机、戒指、手表和挂坠。这种分类适合摆放商品，不适合解释竞争：两副外形相似的眼镜可能是完全不同的产品，而一副眼镜、一只耳机和一枚戒指，也可能在竞争同一件事——谁能最自然地获得用户当下的上下文，理解意图，在不打断现实任务的情况下给出帮助。\n\n",
      "tags": ["AI Wearable","空间计算","多模态交互","产品设计"],
      "lang": "zh-CN"
    },
    {
      "contentId": null,
      "title": "竞品分析为什么不该从参数表开始",
      "url": "https://roy-tong.github.io/notes/competitive-analysis-software-hardware/",
      "date": "2026-08-13T21:35:00+08:00",
      "excerpt": "很多竞品分析都从一张表开始。\n\n",
      "tags": ["竞品分析","软件产品","智能硬件"],
      "lang": null
    },
    {
      "contentId": "ai-native-basic-unit",
      "title": "After AI-Native: The Basic Unit of a Product Has Changed",
      "url": "https://roy-tong.github.io/en/notes/ai-native-basic-unit/",
      "date": "2026-08-13T21:30:00+08:00",
      "excerpt": "Open any traditional editor and you first meet a set of stable basic units. Word processors revolve around pages and paragraphs; presentation software around slides and elements; design tools aroun...",
      "tags": ["AI native","agent","product design"],
      "lang": "en"
    },
    {
      "contentId": "ai-native-basic-unit",
      "title": "AI Native 之后，产品的基本单位变了",
      "url": "https://roy-tong.github.io/notes/ai-native-basic-unit/",
      "date": "2026-08-13T21:30:00+08:00",
      "excerpt": "模型可能已经知道“这是一条核心论点”，但系统里保存的只是“第 3 页左上角的一个文本框”。它可能知道某张图在支撑市场规模，文件里却只记录了图片的位置和尺寸；它可能知道某个镜头属于人物转折，下一次查找仍然只能靠时间码。\n\n",
      "tags": ["AI Native","Agent","产品设计"],
      "lang": "zh-CN"
    },
    {
      "contentId": null,
      "title": "Agent 向左，具身向右：AI 在信息空间与物理世界的分岔",
      "url": "https://roy-tong.github.io/notes/agent-left-embodied-right/",
      "date": "2026-08-13T21:00:00+08:00",
      "excerpt": "一段代码写错了，可以回滚；一封邮件发错了，还能补充说明；数据库被误改，日志和备份能救回来。机械臂碰坏一个物体，损失已经发生；移动机器人撞到人，不能靠“重试”恢复；设备电量耗尽停在危险位置，问题不只是任务失败，还可能引入新的风险。\n\n",
      "tags": ["AI Agent","具身智能","产品判断"],
      "lang": "zh-CN"
    },
    {
      "contentId": null,
      "title": "大疆、影石、安克的组织几乎相反，但都成立",
      "url": "https://roy-tong.github.io/notes/hardware-innovation-organization-management/",
      "date": "2026-08-05T14:30:00+08:00",
      "excerpt": "硬件创新团队如何从少数核心成员出发，逐步建立产品、系统、项目、供应链和规模化经营能力。",
      "tags": ["智能硬件","产品管理","组织管理","产品创新"],
      "lang": null
    },
    {
      "contentId": null,
      "title": "如何从 20 万条反馈中形成可验证的需求判断",
      "url": "https://roy-tong.github.io/notes/scene-user-demand-evidence-research/",
      "date": "2026-08-04T20:00:00+08:00",
      "excerpt": "从一个具体产品决定出发，设计五类证据来源，把每条材料整理成用户、场景、任务、替代方案、摩擦与后果，再用问题、方案、商业行为和反证四条链控制结论。",
      "tags": ["用户研究","产品方法","数据挖掘","社媒调研","AI"],
      "lang": null
    },
    {
      "contentId": null,
      "title": "WAIC 之后：AI 产业开始为“把事做成”买单",
      "url": "https://roy-tong.github.io/notes/waic-from-models-to-systems/",
      "date": "2026-07-23T08:00:00+08:00",
      "excerpt": "WAIC 2026 把一套新的评价标准推到了台前：交付率、连续运行、人工接管，以及完成一次任务的真实成本。",
      "tags": ["AI","WAIC","Agent","具身智能"],
      "lang": null
    },
    {
      "contentId": null,
      "title": "从 AI 应用到真实世界：我的一次转向",
      "url": "https://roy-tong.github.io/notes/from-ai-software-to-physical-world/",
      "date": "2026-07-19T12:00:00+08:00",
      "excerpt": "2025 年 8 月，我停下了 AI 应用创业。此后在 DJI 的大半年，让我重新理解了软件、硬件与竞争门槛，也让我开始把目光投向具身智能和新一代 AI 硬件。",
      "tags": ["AI","具身智能","AI硬件","创业"],
      "lang": null
    },
    {
      "contentId": null,
      "title": "重新开始写作",
      "url": "https://roy-tong.github.io/notes/start-writing/",
      "date": "2026-07-19T08:00:00+08:00",
      "excerpt": "我想把零散的观察公开记录下来，长期整理，也允许后续修正。",
      "tags": ["写作","产品思考"],
      "lang": null
    }
  ]
}
